From 1f9f8ab3f2ea6c8876820ff181f8e1e32652a2bb Mon Sep 17 00:00:00 2001
From: Arjun Bansal <mindscience@gmail.com>
Date: Fri, 1 Sep 2023 15:57:29 -0700
Subject: [PATCH] add metrics

---
 data/compute_metrics.py | 74 +++++++++++++++++++++++++++++++++++++++++
 data/json2l.py          | 17 ++++++++++
 2 files changed, 91 insertions(+)
 create mode 100644 data/compute_metrics.py
 create mode 100644 data/json2l.py

diff --git a/data/compute_metrics.py b/data/compute_metrics.py
new file mode 100644
index 0000000..c5a0c4d
--- /dev/null
+++ b/data/compute_metrics.py
@@ -0,0 +1,74 @@
+import json
+from collections import defaultdict
+
+# Paths to the ground truth and model output JSON files
+ground_truth_path = 'testset-v1.json'
+
+model_output_path = 'gpt-3.5-turbo-testset-v1.json'
+# Accuracy: 0.7107
+# Macro-Averaged Precision: 0.5879
+# Macro-Averaged Recall: 0.5736
+# Macro-Averaged F1 Score: 0.5755
+
+#model_output_path = 'pandalm-7b-testset-v1.json'
+# Accuracy: 0.6677
+# Macro-Averaged Precision: 0.5738
+# Macro-Averaged Recall: 0.5750
+# Macro-Averaged F1 Score: 0.5743
+
+def get_majority_vote(*votes):
+    vote_count = defaultdict(int)
+    for vote in votes:
+        vote_count[vote] += 1
+    return max(vote_count, key=vote_count.get)
+
+def compute_metrics(ground_truth_path, model_output_path):
+    with open(ground_truth_path, 'r') as gt_file, open(model_output_path, 'r') as mo_file:
+        ground_truths = json.load(gt_file)
+        model_outputs = json.load(mo_file)
+
+        confusion_matrix = defaultdict(lambda: defaultdict(int))
+
+        for gt, mo in zip(ground_truths, model_outputs):
+            majority_vote = get_majority_vote(gt['annotator1'], gt['annotator2'], gt['annotator3'])
+            if 'gpt' in model_output_path:
+                model_prediction = mo['gpt_result']
+            elif 'panda' in model_output_path:
+                model_prediction = mo['pandalm_result']
+            if model_prediction == "Tie" or model_prediction == "tie" or model_prediction =="garbage":
+                model_prediction = 0
+            else:
+                model_prediction = int(model_prediction)
+
+            confusion_matrix[majority_vote][model_prediction] += 1
+
+        # Accuracy calculation
+        possible_classes = [0, 1, 2]
+        total_true_positives = sum(confusion_matrix[i][i] for i in possible_classes)
+        total_instances = sum(sum(confusion_matrix[i].values()) for i in possible_classes)
+        accuracy = total_true_positives / total_instances
+
+        metrics = {}
+        
+        for cls in possible_classes:
+            TP = confusion_matrix[cls][cls]
+            FP = sum(confusion_matrix[x][cls] for x in possible_classes) - TP
+            FN = sum(confusion_matrix[cls][x] for x in possible_classes) - TP
+            precision = TP / (TP + FP) if (TP + FP) != 0 else 0
+            recall = TP / (TP + FN) if (TP + FN) != 0 else 0
+            f1 = (2 * precision * recall) / (precision + recall) if (precision + recall) != 0 else 0
+
+            metrics[cls] = {'precision': precision, 'recall': recall, 'f1': f1}
+
+        macro_avg = {}
+        for metric in ['precision', 'recall', 'f1']:
+            macro_avg[metric] = sum(metrics[cls][metric] for cls in possible_classes) / 3
+
+        return accuracy, macro_avg
+
+
+accuracy, macro_avg = compute_metrics(ground_truth_path, model_output_path)
+print(f"Accuracy: {accuracy:.4f}")
+print(f"Macro-Averaged Precision: {macro_avg['precision']:.4f}")
+print(f"Macro-Averaged Recall: {macro_avg['recall']:.4f}")
+print(f"Macro-Averaged F1 Score: {macro_avg['f1']:.4f}")
diff --git a/data/json2l.py b/data/json2l.py
new file mode 100644
index 0000000..273fb99
--- /dev/null
+++ b/data/json2l.py
@@ -0,0 +1,17 @@
+import json
+
+def convert_json_to_jsonl(json_file_path, jsonl_file_path):
+    with open(json_file_path, 'r') as json_file:
+        data = json.load(json_file)
+        
+        if not isinstance(data, list):
+            raise ValueError("The JSON file should contain a list of objects")
+        
+        with open(jsonl_file_path, 'w') as jsonl_file:
+            for item in data:
+                jsonl_file.write(json.dumps(item) + '\n')
+                
+    print(f"Converted {json_file_path} to {jsonl_file_path}")
+
+# Example usage:
+convert_json_to_jsonl('train.json', 'train.jsonl')